Reading Time: 9 minutes

Euro-era Croatia is united in money, and still divided in how jobs respond to growth.

1. One capital, many labour markets: What the county evidence says

Zoomed out, Okun’s Law is a comforting regularity: when the economy runs hot, unemployment cools; when output contracts, joblessness climbs. Zoomed in to Croatia’s counties, it becomes less a “law” than a set of local habits, some sturdy, some sluggish, some startlingly asymmetric.

Part I’s graphics already delivered the broad intuition. In many counties, cyclical slack in output and cyclical slack in unemployment co-move in the expected opposite directions, with downturn episodes pulling the relationship into sharper focus. But the same pictures also made a second point unavoidable: counties are not merely smaller Croatias. They differ in baseline unemployment, volatility, exposure to shocks, and, crucially for Okun, how quickly labour markets respond when output turns.

Part II then did the necessary work that pictures alone cannot do. It asked whether the county panel behaves in a way that makes inference legitimate: whether the series are stationary or drifting, whether counties share common shocks, whether slopes are plausibly common or heterogeneous, whether a long-run relationship exists across counties, and whether short-run dynamics look symmetric or not. That pipeline matters because the county panel is exactly the setting where naïve inference is easiest and most dangerous: common shocks can make relationships look strong even when they are misspecified, while genuine heterogeneity can be flattened into an “average Croatia” that describes no county particularly well.

Put simply, the combined evidence says this: Croatia’s county panel supports an Okun relationship, especially in the gap specification; it also insists that the relationship is interdependent across counties, structurally unstable around major episodes, and potentially asymmetric in how labour markets respond to expansions versus contractions.

2. What the combined evidence actually says about Okun in Croatia’s regions

The report’s results form a coherent story once you treat them as layers rather than as competing verdicts.

First layer: The data are not “well-behaved” in the same way across specifications.
In levels, county LGDP and unemployment rates show persistence that requires careful handling. That is why the report begins with panel unit-root testing and why it revisits those conclusions once cross-sectional dependence is acknowledged. First-generation tests provide a familiar I(1)–I(0) split, levels drifting, first differences stationary, while the gap variables are more plausibly stationary cycles (the point of extracting slack measures). The gap lens is not just a stylistic preference; it is structurally aligned with the properties the report finds (Hodrick & Prescott, 1997).

Second layer: Counties move together, so the toolkit must admit it.
The cross-sectional dependence results are a polite way of saying that a county panel is not twenty independent experiments. Counties share national policy, common external shocks, and integrated markets. The report therefore treats dependence testing as a bridge to second-generation methods rather than as a decorative diagnostic (Pesaran, 2004; Pesaran & Xie, 2021; Juodis & Reese, 2022). Once dependence is present, it is not enough to rely on tests that assume independence; the analysis must move to methods that explicitly allow for shared factors.

That shift matters economically. It says that part of the Okun channel is a national cycle passing through regional labour markets, even if local structures shape the amplitude and speed. In other words, there is a “Croatia-wide” macro rhythm; counties play different instruments.

Third layer: “One Croatia” is an assumption the data do not always support.
Slope homogeneity testing is the report’s way of asking whether it is defensible to talk about one Okun coefficient for all counties. The results are not uniformly simple, some variants reject homogeneity more strongly than others, but the direction is clear enough: heterogeneity is not safely ignorable (Pesaran & Yamagata, 2008). That has an obvious interpretation. The unemployment response to a given output movement can differ across counties because sectoral structure, labour-market frictions, and adjustment margins differ across space.

This is also where the county frame becomes politically relevant. A national narrative that treats labour-market adjustment as uniform risks misreading the local distribution of pain and recovery. If some counties have flatter Okun slopes or slower adjustment, then “growth is back” can coexist with persistent local slack.

Fourth layer: Breaks are not a nuisance; they are the plot.
The report’s structural-break work is a reminder that the sample period contains episodes large enough to rewire relationships, not merely to perturb them. Multiple-break logic indicates structural instability in key relationships, with breakpoints clustering around the familiar crisis era and around the pandemic period (Bai & Perron, 1998). The report reinforces this theme with break-aware panel unit-root testing, which shows that stationarity conclusions can change depending on whether breaks are treated as unknown and endogenously located or imposed mechanically at specific crisis dates (Chen, Karavias, & Tzavalis, 2022).

The economic meaning is straightforward: the Okun relationship is not guaranteed to be stable across regimes. County labour markets can adjust differently before and after large disruptions; the same output change can map into different unemployment outcomes depending on the period’s institutional and structural context.

Fifth layer: Long-run linkage appears strongest in slack terms.
Panel cointegration results, read in the round across the report’s test families, support a long-run relationship most convincingly in the gap specification. That is, county output gaps and unemployment gaps are more consistently tied together by a mean-reverting long-run linkage than the level variables are under the same breadth of assumptions (Pedroni, 1999; Kao, 1999; Westerlund, 2007). This does not mean levels are irrelevant. It means the gap relationship is more empirically coherent for the kind of cyclical policy questions that readers actually ask: is the economy operating below its local potential, and is that slack showing up in county unemployment?

Sixth layer: Dynamics and asymmetry complicate the “simple slope” story.
The report’s ARDL results align with a familiar macro reality: labour-market adjustment tends to be gradual, with error-correction dynamics indicating convergence rather than instant translation. The use of bounds-style ARDL logic and dynamic heterogeneous panel ideas is motivated precisely by that separation of short-run movement from long-run equilibrium (Pesaran, Shin, & Smith, 2001; Pesaran & Smith, 1995).

Then come the nonlinear results, which are where Okun becomes most politically legible. The NARDL evidence supports the idea that the labour market’s response can differ across positive and negative output movements, contractions can widen slack quickly, recoveries can narrow it more slowly or unevenly. That is not merely a statistical curiosity; it is the difference between a recession that scars local labour markets and a recovery that leaves uneven footprints (Shin, Yu, & Greenwood-Nimmo, 2014).

Seventh layer: Predictability runs largely from output to unemployment, while slack can feed back.
The panel causality results, interpreted carefully as predictability rather than metaphysical cause, reinforce Okun’s directional logic in change terms, output movements help predict unemployment movements more clearly than the reverse. In slack terms, the report finds stronger interaction and potential bidirectionality, consistent with the idea that once cyclical slack is present, it can feed back through demand, confidence, and local adjustment dynamics. Where the report uses improved causality tests suited to fixed-T panels, the interpretive thrust is that the output side contains meaningful leading information for labour outcomes, while slack can become a two-way system once established (Xiao, Juodis, Karavias, & Sarafidis, 2023).

3. Where the Okun channel looks strongest, and where it leaks across space

If you had to translate the report into one sentence that avoids both exaggeration and understatement, it would be: Croatia’s county Okun relationship is strongest when expressed as cyclical slack, and weakest when you insist on uniformity, stability, and symmetry.

The report’s strongest empirical footing is the gap model: output slack and unemployment slack move together across counties with a long-run linkage that survives a range of tests. That is also where the economic interpretation is cleanest. A county that is below trend in output tends to be above trend in unemployment; when the output gap closes, unemployment slack tends to narrow, but not instantly.

Where does it “leak”? The report’s own pipeline identifies the pressure points.

Leak #1: common shocks plus heterogeneity.
Cross-sectional dependence means counties share shocks; slope heterogeneity means counties respond differently. Put those together and a national average becomes simultaneously informative and misleading: informative about the shared cycle, misleading about local adjustment.

Leak #2: regime instability.
Breaks imply that a coefficient estimated across the full sample is an average across regimes. That average may be a poor guide in the very moments policymakers care about most, around major disruptions. Structural instability is not a technical nuisance; it is a warning against treating the Okun coefficient as a fixed policy instrument.

Leak #3: asymmetric adjustment.
If negative shocks widen unemployment slack faster than positive shocks narrow it, then “recoveries” can be statistically real and socially incomplete. That asymmetry is particularly relevant at the county level because local labour markets can carry inertia differently, through sectoral concentration, migration, and matching frictions.

In short, the report’s county panel does not reject Okun’s Law. It rejects the simplistic version of it, the one where growth buys jobs at a stable exchange rate, county by county, quarter by quarter, as if labour markets were frictionless and history never interrupted.

4. EU membership, euro adoption (2023), and the constraints of credibility

Croatia’s EU membership provides an institutional backdrop: broad alignment with EU frameworks and a policy environment shaped by integration. The report also notes euro adoption in 2023 as a meaningful macro regime marker, an anchor for credibility and a constraint on certain domestic stabilisation levers.

The key point, consistent with the report’s county evidence, is that a stronger macro anchor does not automatically equal uniform regional transmission. If counties differ in Okun slopes, adjustment speeds, and asymmetries, then the “same” national or euro-area macro environment can still yield different labour-market experiences across space.

This is where the county lens is most useful for euro-era Croatia. Under a credibility anchor, the macro cycle may be more constrained in certain dimensions, but the distribution of adjustment can still vary regionally. The report’s emphasis on heterogeneity and dependence suggests a practical interpretation: national and euro-area conditions shape the overall rhythm, while county structures shape how strongly and how quickly labour markets dance to it.

One cautious, report-consistent inference follows: if euro adoption strengthens the macro anchor, then the policy frontier shifts toward internal adjustment, labour-market functioning, matching efficiency, regional mobility, and targeted measures where inertia is strongest. The county panel does not ask Croatia to abandon national narratives. It asks Croatia to stop pretending those narratives land evenly.

5. Practical takeaways

The report’s results imply that Croatia’s county Okun channel is real, but conditional. That leads to a policy stance that is less slogan-friendly and more useful.

• Use slack measures for the macro conversation, and county heterogeneity for the policy design. Gap-based Okun evidence is empirically stronger and conceptually closer to stabilisation questions; county differences matter for where and how labour-market support is targeted (Hodrick & Prescott, 1997; Pedroni, 1999; Westerlund, 2007).
• Treat downturn prevention as employment policy. If the relationship is asymmetric, recessions can do disproportionate labour-market damage, and recoveries may not symmetrically repair it (Shin et al., 2014).
• Avoid “average Croatia” complacency. Cross-sectional dependence means a shared cycle; heterogeneity means uneven transmission. A single national coefficient is a summary, not a map (Pesaran, 2004; Pesaran & Yamagata, 2008).
• Make structural change a standing assumption, not an emergency footnote. Breaks are empirically present and economically plausible; policies should be robust to regime shifts rather than calibrated to a single historical average (Bai & Perron, 1998; Chen et al., 2022; Ditzen, Karavias, & Westerlund, 2024).

6. What to watch next

The report ends up implying a useful monitoring agenda, what would tell us whether Croatia’s county labour markets are converging toward a more uniform Okun channel, or diverging into more persistent regional asymmetry.

• Do county adjustment speeds converge? If error-correction dynamics become faster and more uniform across counties, that would signal improved labour-market responsiveness in the euro era (Pesaran & Smith, 1995; Pesaran et al., 2001).
• Does asymmetry persist in the next downturn? If negative shocks continue to widen unemployment slack faster than recoveries narrow it, that is a policy problem, not just a modelling detail (Shin et al., 2014).
• Do new breakpoints appear post-2023? The report’s emphasis on structural instability suggests that the euro regime may coincide with altered relationships, but that must be inferred from data rather than assumed (Bai & Perron, 1998; Ditzen et al., 2024).
• Does predictability remain output-led? If output continues to lead unemployment movements in change terms and interacts strongly in slack terms, that supports the policy value of cyclical surveillance and early stabilisation (Xiao et al., 2023).

A final thought, in the spirit of a slightly punchy Economist ending: Croatia has joined a currency union; it has not joined a single labour market. The county panel evidence does not deny national progress, it simply insists that national progress is not evenly distributed, and that the growth–jobs link has local accents. Zagreb may write the headline, but the counties supply the footnotes.

7. References

Bai, J., & Perron, P. (1998). Estimating and testing linear models with multiple structural changes. Econometrica, 66(1), 47–78. DOI: https://doi.org/10.2307/2998540

Chen, P., Karavias, Y., & Tzavalis, E. (2022). Panel unit-root tests with structural breaks. The Stata Journal, 22(3), 664-678. DOI: https://doi.org/10.1177/1536867X221124541

Ditzen, J., Karavias, Y., & Westerlund, J. (2024). Sequential estimation of multiple breaks in time series and panel data. Journal of Applied Econometrics, 40(1), 74-88. DOI: https://doi.org/10.1002/jae.3097

Hodrick, R. J., & Prescott, E. C. (1997). Postwar U.S. business cycles: An empirical investigation. Journal of Money, Credit, and Banking, 29(1), 1–16. DOI: https://doi.org/10.2307/2953682

Johansen, S. (1991). Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models. Econometrica, 59(6), 1551–1580. DOI: https://doi.org/10.2307/2938278

Juodis, A., & Reese, S. (2022). The incidental parameters problem in testing for remaining cross-section correlation. Journal of Business Economics and Statistics, 40(3), 1191-1203. DOI: https://doi.org/10.1080/07350015.2021.1906687

Kao, C. (1999). Spurious regression and residual-based tests for cointegration in panel data. Journal of Econometrics, 90, 1-44. DOI: https://doi.org/10.1016/S0304-4076(98)00023-2

Okun, A. M. (1962). Potential GNP: Its measurement and significance. American Statistical Association, Proceedings of the Business and Economic Statistics Section, 98–104.

Pedroni, P. (1999). Critical values for cointegration tests in heterogeneous panels with multiple regressors. Oxford Bulletin of Economics and Statistics, 61, 653-70. DOI: https://doi.org/10.1111/1468-0084.0610s1653

Pesaran, M. H. (2004). General diagnostic tests for cross-section dependence in panels. CESifo Working Paper Series No. 1229.

Pesaran, M. H. (2007). A simple panel unit root test in the presence of cross-section dependence. Journal of Applied Econometrics, 22(2), 265–312. DOI: https://doi.org/10.1002/jae.951

Pesaran, M. H., & Smith, R. P. (1995). Estimating long-run relationships from dynamic heterogeneous panels. Journal of Econometrics, 68(1), 79-113. DOI: https://doi.org/10.1016/0304-4076(94)01644-F

Pesaran, M. H., & Xie, Y. (2021). A Bias-corrected CD test for error cross-sectional dependence in panel data models with latent factors. Cambridge Working Papers in Economics 2158.

Pesaran, M. H., & Yamagata, T. (2008). Testing slope homogeneity in large panels. Journal of Econometrics, 142(1), 50–93. DOI: https://doi.org/10.1016/j.jeconom.2007.05.010

Pesaran, M. H., Shin, Y., & Smith, R. J. (2001). Bounds testing approaches to the analysis of level relationships. Journal of Applied Econometrics, 16(3), 289–326. https://doi.org/10.1002/jae.616

Shin, Y., Yu, B., & Greenwood-Nimmo, M. (2014). Modelling asymmetric cointegration and dynamic multipliers in a nonlinear ARDL framework. In Festschrift in Honor of Peter Schmidt (pp. 281–314). Springer. DOI: https://doi.org/10.1007/978-1-4899-8008-3_9

Westerlund, J. (2007). Testing for error correction in panel data. Oxford Bulletin of Economics and Statistics, 69(6), 709–748. DOI: https://doi.org/10.1111/j.1468-0084.2007.00477.x

Xiao, J., Juodis, A., Karavias, Y., & Sarafidis, V. (2023). Improved tests for Granger causality in panel data. The Stata Journal, 23(1), 230-242. DOI: https://doi.org/10.1177/1536867X231162034